Built a Tamil abusive text detection model using IndicBERTv2 🧠. Applied domain adaptation by fine-tuning the MLM model on unlabeled Tamil text before supervised classification. Achieved 96.8% accuracy and 96.7% F1 score 📊. A great hands-on experience in transfer learning and regional language NLP 🚀.